[gregorio]

Thanks. I understand it better now. Basically, I can record "colder" than before and not worry about the possibility of clipping. This is definitely worthwhile. Can't tell you how often a perfectly good take was almost ruined because of a few unexpectedly loud hits here and there.

Yep, castleofargh's response was spot on. And yes, that's exactly the benefit of 24bit for recording, no intrinsic quality benefit, just a lot more headroom to play with. Of course, that's only the recording medium signal to noise ratio you don't really have to worry about with 24bit, you've still got mic, environmental noise/noise-floor and pre-amp noise to worry about!

G

 

[Dobrescu George]

So.... Most bands won't get into recording into 24 bit any time soon because it won't become an ISO standard?...

 

[71 dB]

In music production you want dynamic headroom as mentioned before and all digital effects increase noise floor. Say you filter frequencies below 50 Hz out on one track and quantization noise is added at the level of least significant bit. Everytime you do something noise is added and it cumulates. So, you want headroom in the lower end too to be able to drop the noisy least significant bits away when downsampling to 16 bit for CD.

 

[pinnahertz]

In music production you want dynamic headroom as mentioned before and all digital effects increase noise floor. Say you filter frequencies below 50 Hz out on one track and quantization noise is added at the level of least significant bit. Everytime you do something noise is added and it cumulates. So, you want headroom in the lower end too to be able to drop the noisy least significant bits away when downsampling to 16 bit for CD.

While true in theory, in practice there are 4 to 6 LSB worth of noise already because there ain't but one ADC in the world capable of better than 20 bits of noise performance, even if it's a 24 bit ADC. Add a mic preamp in front of the ADC, and it's even worse. I doubt the low level noise added by any effect would change what's already there much at all. Especially since internal DAW processing is already at 64 bit fp.

 

[pinnahertz]

So.... Most bands won't get into recording into 24 bit any time soon because it won't become an ISO standard?...

I'm not sure what "most bands" means, but the default today is to record at 24 bits...because you can, and it's useful in production to do so. All DAWs are up to 32 or 64 bit floating point internally.

 

[bigshot]

I've never done anything above 24/96 though. Not much reason to. It can only cause problems.

 

[pinnahertz]

I've never done anything above 24/96 though. Not much reason to. It can only cause problems.

You should try 25/97, it's just a bit better.

 

[71 dB]

While true in theory, in practice there are 4 to 6 LSB worth of noise already because there ain't but one ADC in the world capable of better than 20 bits of noise performance, even if it's a 24 bit ADC. Add a mic preamp in front of the ADC, and it's even worse. I doubt the low level noise added by any effect would change what's already there much at all. Especially since internal DAW processing is already at 64 bit fp.

Yeah, of course. 24 bits means the additional noise happens at 48 dB lower level. Prosessing can be at 64 bit floating point, but in the end you need to truncate + dither it to 24 bits.

 

[gregorio]

[1] Yeah, of course. 24 bits means the additional noise happens at 48 dB lower level. [2] Prosessing can be at 64 bit floating point, but in the end you need to truncate + dither it to 24 bits.

1. No it doesn't. When processing, 24bit or 16bit makes no difference because the processing occurs in a 64bit processing/mixing environment and the noise added from processing in the LSB is many, many hundreds of dB below audibility. Even with 1,000 channels of audio and multiple processors on each, still the noise would be way below audibility! If I remember correctly (and I may not as it was about 15 years ago) even with the older 48bit fixed mixing environment, you needed 277 processors for the resultant quantisation noise to reach -120dB.

2. Noise from truncating 64bit float to 24bit fixed would average the LSB, IE. Peak average of truncation noise would be at -138dB. As this is unresolvable, no one I know bothers dithering to 24bit. Even dithering to 16bit is only a standard procedure on the basis of "better safe than sorry", rather than because the resultant error is likely to be audible.

So.... Most bands won't get into recording into 24 bit any time soon because it won't become an ISO standard?...

I can understand you not wanting to read the entire thread before posting but at least the OP and the previous handful of posts would be advisable!

G

 

[71 dB]

1. No it doesn't. When processing, 24bit or 16bit makes no difference because the processing occurs in a 64bit processing/mixing environment and the noise added from processing in the LSB is many, many hundreds of dB below audibility. Even with 1,000 channels of audio and multiple processors on each, still the noise would be way below audibility! If I remember correctly (and I may not as it was about 15 years ago) even with the older 48bit fixed mixing environment, you needed 277 processors for the resultant quantisation noise to reach -120dB.

I am talking about the the noise cumulation when several 64 bit processing are done. After every processing you truncate + dither the result and that's when 16 bit or 24 bit makes a big difference. At 16 bit the noise cumulates fast above -90 dBFS* , while at 24 bit the noise level is perhaps -110 dBFS, the original noise from mic amps etc. because the cumulated noise add the noise power so little.

* Let's assume you have 10 tracks at 16 bits truncated (no dither) to from 24 bit. Your quantization noise for each track is at level -98 dBFS. When you mix those tracks together, the noise power gets 10-fold and is at level -88 dBFS. If you have 100 tracks, the end result noise is at level -78 dBFS, which starts to be bad.

2. Noise from truncating 64bit float to 24bit fixed would average the LSB, IE. Peak average of truncation noise would be at -138dB. As this is unresolvable, no one I know bothers dithering to 24bit. Even dithering to 16bit is only a standard procedure on the basis of "better safe than sorry", rather than because the resultant error is likely to be audible.

The dynamic range at 24 bit without dither is 20 * log (2^24 * sqrt(3/2)) = 146.26 dB. Yes, 24 bit is fine without dither, but every CD use dither, which extends usable dynamic range 10-20 dBs.

Are you saying 16 bits is just fine in studio, because prosessing is done at 64 bit? My understanding is that 24 bits allow nice headroom and no more than 32 bits is needed in prosessing. For consumer audio 16bit / 44.1 kHz is enough.

 

[gregorio]

[1] I am talking about the the noise cumulation when several 64 bit processing are done. [2] After every processing you truncate + dither the result and that's when 16 bit or 24 bit makes a big difference.

1. So am I or rather, I'm talking about many hundreds of 64bit processes being done!

2. No you do not truncate + dither after every processing and that's why there is no difference between 16 and 24 bit! The entire mixing process is carried out in the 64bit mixing environment. You don't perform a process on a track, bounce the track back to a 16 or 24 bit file (with truncation or dither), then apply some other process at 64bit and bounce/truncate it back to 16 or 24 bit again, ad infinitum until you finish the mix. If you did, it would take forever to create a mix and would end up, as you suggest, with significant noise. You would get significant noise even if you followed this process with 24 bit, which is why this is not how mixing works and why no mixing environments are 24 bit!! What actually happens is that your 16 or 24bit tracks are loaded into a 64bit mixing environment. You perform a process and the result stays in the mixer (in RAM at 64bit), you perform another process (or numerous other processes) at 64bit using the 64bit result from the last process and everything stays at 64bit all the time, including the truncation error. The only point at which there is any truncation above the 64th bit is when the mixing is complete and you bounce that completed mix out of the mixing environment to say an audio file (in 16 or 24bit). Therefore, there is no accumulation of 16 or 24bit truncation noise, all accumulated truncation noise occurs at 64bit and is completely inaudible even with many hundreds of tracks and processors!

The rest of your post is therefore irrelevant/incorrect!

G

 

[71 dB]

1. So am I or rather, I'm talking about many hundreds of 64bit processes being done!

2. No you do not truncate + dither after every processing and that's why there is no difference between 16 and 24 bit! The entire mixing process is carried out in the 64bit mixing environment. You don't perform a process on a track, bounce the track back to a 16 or 24 bit file (with truncation or dither), then apply some other process at 64bit and bounce/truncate it back to 16 or 24 bit again, ad infinitum until you finish the mix. If you did, it would take forever to create a mix and would end up, as you suggest, with significant noise. You would get significant noise even if you followed this process with 24 bit, which is why this is not how mixing works and why no mixing environments are 24 bit!! What actually happens is that your 16 or 24bit tracks are loaded into a 64bit mixing environment. You perform a process and the result stays in the mixer (in RAM at 64bit), you perform another process (or numerous other processes) at 64bit using the 64bit result from the last process and everything stays at 64bit all the time, including the truncation error. The only point at which there is any truncation above the 64th bit is when the mixing is complete and you bounce that completed mix out of the mixing environment to say an audio file (in 16 or 24bit). Therefore, there is no accumulation of 16 or 24bit truncation noise, all accumulated truncation noise occurs at 64bit and is completely inaudible even with many hundreds of tracks and processors!

The rest of your post is therefore irrelevant/incorrect!

G

Okay, I see. I didn't know that because I work with Audacity. Millionaires can work with their 64 bit monsters… …thanks for the lecture.

 

[gregorio]

Okay, I see. I didn't know that because I work with Audacity. Millionaires can work with their 64 bit monsters… …thanks for the lecture.

Logic Pro = $199.99
Pro Tools = $199 (annual subscription).
Reaper = $60

You don't need to be a millionaire! And 64bit isn't a monster, my iPhone is 64bit and cost more than these DAWs!

G

 

[pinnahertz]

Logic Pro = $199.99
Pro Tools = $199 (annual subscription).
Reaper = $60

You don't need to be a millionaire! And 64bit isn't a monster, my iPhone is 64bit and cost more than these DAWs!

G

I find Audacity a useful tool on occasion, but too limiting for real work. Great tool for free, but hardly an industry standard. G, you'll be interested to know that when I needed deliberate clipping Audacity was the tool that did it, but I had to set it to 16 bit (it also does 32 bit float).

 

[gregorio]

G, you'll be interested to know that when I needed deliberate clipping Audacity was the tool that did it, but I had to set it to 16 bit (it also does 32 bit float).

Yep, taking it out and into another system/DAW is how I solved the problem at the time. I could of course just have looped back and clipped the ADC/DAC. I can also achieve it with a couple of the plugins I now own, which specifically provide an option to create digital clipping and various other sorts of overload distortion. This is just between you and me though!

The point I was making is that the quoted statement was nonsense, it's impossible to accidentally clip a pro DAW and without specific tools or leaving the DAW it's even impossible (AFAIK) to clip deliberately!

G